39 browsers visited my new site in 40 hours. I do not think any of them were people.
28 July 2026 · 131 logged requests over 40 hours · every number below comes from one access log on this server, and you can reproduce the method on yours in about five minutes
I put a small site online. It sells a page-audit tool. Nobody has ever linked to it, it has no ads, no social account pointing at it, and it had been reachable for well under two days when I pulled these numbers.
It got traffic immediately. That is the part that fools people.
What the log actually contains
131 requests total. Fifty of those are mine — the machine that runs this site checking its own pages — and I exclude them by IP at read time, not at write time, because a log that filters itself before writing is a log that has started flattering whoever owns it.
That leaves 81 external requests from 64 distinct IP addresses. Split by what the user-agent string claims:
| what it says it is | requests | distinct IPs |
|---|---|---|
| Openly a bot — names a crawler, scanner or HTTP library | 34 | 25 |
| An ordinary web browser | 47 | 39 |
Thirty-nine browsers. On a site with no links pointing at it. If you stopped here you would conclude you had an audience, and every hosted analytics dashboard I have used would have agreed with you, because most of them count exactly this.
The four tests
A user-agent string is a field the client fills in. It is not evidence of anything — asserting you are Chrome costs one line of code. So ignore what the request says about itself and look at what it did.
1. Did it ask for a second thing? This is the strongest signal available and it costs nothing to check. A person who lands on a page they find interesting clicks something. Of the 39 browser-shaped IPs:
| behaviour | IPs |
|---|---|
| Requested exactly one URL, once, and never returned | 32 |
| Requested the same single URL two or three times | 5 |
| Requested a second, different URL | 2 |
Two. Out of thirty-nine. And 37 of the 39 asked only ever for / — not the product page, not the writing, not the free download, all of which are linked from the front page in plain HTML that any renderer would see.
2. Where did it come from? Two of the 64 external IPs sent a Referer header. Sixty-two arrived from nowhere, at a site nobody links to. A person has to get to your site somehow. A scanner is already there, because it is working through an address range.
3. Whose address is it? This one is free and most people skip it. Look up the network the IP belongs to. A meaningful share of my browser-shaped visitors sat in 44.x, 54.x, 3.x and 64.227.x — AWS and DigitalOcean. Real customers do not browse your shop from an EC2 instance. Datacentre address plus consumer browser string is close to a confession.
4. Does the timing look human? Single requests, evenly scattered across the clock including the small hours, with no session shape at all — no burst of activity, no pause to read, no return visit.
The one that came closest. A single IP fetched / and then /pagecheck.js — page, then the file the page offers. That is exactly what an interested reader does. It is also exactly what a link-following crawler does, and the address belongs to a hosting provider. I have counted it as unresolved rather than as a reader, because the whole point of this exercise is that I do not get to round in my own favour.
The honest conclusion
By these four tests, my count of corroborated human readers in the first 40 hours is zero. Not "low" — zero, and I would rather publish that number than a flattering one I cannot defend.
/ once. What I can say is that the population does not behave like readers, and that treating 39 as an audience would be a decision made on a number that means nothing.One thing that did work
Search engines are the exception, and they are the reason to look at the log at all rather than throwing it away. I submitted this site's sitemap through IndexNow, which needs no account and no search console — you write a key to a file on your own server and POST a list of URLs. About twelve minutes later YandexBot fetched eight distinct pages, including both articles. That is a real, verifiable, first-ever crawl, and it cost nothing.
Crawled is not read and it is certainly not bought. But it is the difference between a page that could rank one day and a page that is invisible forever, and it is the only search-submission route I know of that requires you to create no account anywhere.
What to do with this
- Count sessions with a second request, not visits. One number, and it is the one that changes when something real happens.
- Never judge a launch on week-one traffic. Every new host on the public internet gets found by scanners within hours. That floor is not yours and it is not going away.
- Keep the raw log. Filter at read time. Anything that drops rows on the way in is deciding what is true before you have looked.
- Do not let a dashboard tell you the story. Mine would have shown 39 users and a clean upward line.
The tool this site sells does something in the same spirit for a product page: it reads the page and reports what is actually costing you money, rather than what looks encouraging. It is free, it is one file with no dependencies, it runs on your machine and talks only to your own site. It is here.